Please use this identifier to cite or link to this item:
Title: Real-Time Event-Driven Learning in Highly Volatile Systems: A case for embedded Machine Learning for SCADA systems
Authors: Gonçalves, M
Sousa, P
Mendes, J
Danishvar, M
Mousavi, A
Issue Date: 9-May-2022
Publisher: IEEE
Citation: Gonçalves, M., Sousa, P., Mendes, J., Danishvar, M. and Mousavi, A. (2022) 'Real-Time Event-Driven Learning in Highly Volatile Systems: A case for embedded Machine Learning for SCADA systems', IEEE Access, 10, pp. 50794 - 50806 (13). doi: 10.1109/ACCESS.2022.3173376.
Abstract: Copyright © The Author(s) 2022. Extracting key system parameters and their impact on state transition is a necessity for knowledge and data engineering. In Decision Support Systems, the quest for yet more efficient and faster methods of sensitivity analysis (SA) and feature extraction in complex and volatile systems persists. A new improved event tracking methodology, the fastTracker, for real-time SA in large scale complex systems is proposed in this paper. The main feature of fastTracker is its high-frequency analytics using meager computational cost. It is suitable for data processing and prioritization in embedded systems, Internet of Things (IoT), distributed computing (e.g. Edge computing) applications. The presented algorithm’s underpinning rationale is event driven; its objective is to correctly and succinctly quantify the sensitivity of observable changes in the system (output) with respect to the input variables. To demonstrate the performance of the proposed fastTracker methodology, fastTracker was deployed in the Supervisory control and data acquisition (SCADA) system from real cement industry. fastTracker has been verified by system experts in real industrial application. Its performance was compared with other real-time event-based SA techniques. The comparison revealed savings of 98.8% in processing time per sensitivity index and 20% in memory usage when compared with EventTracker, its closest rival. The proposed methodology is more accurate and 80.9% faster than an entropy-based method. Its application is recommended for reinforced learning and/or formulating system key performance indicators from raw data.
Appears in Collections:Dept of Mechanical and Aerospace Engineering Research Papers

Files in This Item:
File Description SizeFormat 
FullText.pdf6.71 MBAdobe PDFView/Open

This item is licensed under a Creative Commons License Creative Commons